Anthropic takes Claude API cache diagnostics out of beta

Anthropic has moved cache diagnostics on the Claude API out of beta, dropping the special header developers previously needed to turn the feature on.

What's new

The change is a straightforward graduation from beta to standard availability. Anthropic's release notes state it plainly: "Cache diagnostics is out of beta on the Claude API and no longer requires the cache-diagnosis-2026-04-07 beta header." Developers who want diagnostic data on their prompt cache behavior now include a diagnostics object directly on a Messages request to opt in, with no beta flag required.

On the response side, the behavior has also changed to be always-present rather than conditional: calls to POST /v1/messages now always include a diagnostics field in the response, returning null when a caller didn't request it rather than omitting the field entirely. That makes the response shape predictable for client code parsing it, instead of having to branch on whether the field exists.

Context

Cache diagnostics gives developers visibility into how Anthropic's prompt caching is actually behaving on their requests, useful for debugging unexpected cache misses or verifying that a caching strategy is delivering the expected cost and latency savings. It had been gated behind the cache-diagnosis-2026-04-07 beta header since introduction; removing that gate is a standard signal that Anthropic considers the feature stable enough for general use without an opt-in flag.

Why it matters

This is a small but practical change for any team that has built cost or latency monitoring around Claude's prompt caching, since it removes a header dependency from client code and standardizes the response shape. It also continues a pattern visible across Anthropic's recent release notes, where the company is steadily hardening caching-related tooling on the API as prompt caching becomes a bigger lever for cost control on long-context, agentic workloads.